Providing screenshot as a tool for the agent has massively increased the visual accuracy of the code changes that it makes. When providing Figma design screenshots, we've seen the agent be very precise about the changes it makes and iterate until perfection 🔎

For Vite + React projects (including Remix!), the "provide context with React inspector" tool extracts the exact code range of the component that maps to the DOM node. First-class support for Next.js is ships next week too ✨

This one is for the devs building frontend applications 🥁🥁🥁 Aide now ships with a browser and React Devtools built-in! Attach screenshots & code just by clicking around your app 📸 ...and, the agent *looks* at the browser every few steps to visually verify its work! 👀

📸 Sidecar v0.1.14 - Early alpha support for o1-mini to guide sonnet3.5 We are excited to announce Agentic reasoning. We heard your feedback about our agents going off-track and doing too much, we have an objective answer to it

> be you > get $20 subscription > get unlimited and full context access to o1-mini, sonnet3.5(new), deepseek v3, gemini 1.5 pro > profit?

what if the extension was a MCP server

SWE-agent right in your editor. We are launching a new extension which provides the best of our agents in your editor of choice. It can: - read directory files - run terminal commands - edit files - and more... Here is our SWE-agent editing the website itself 😉

105M tokens/day @openrouter

you wanted support for any and all models, its here Testing out the sonnet via @openrouter Cmd+Shift+P: Model Selection (JSON) to any model!

We are improving the UX for probing: - We now show the results of the probe directly in the editor - Decorations are improved so you can see where the agent decided to click and the symbol to follow

And the developer stays in the loop for all the actions performed by the agents, fully able to dive in and understand what the agent is up to.

A single agent is scoped to work on a code-symbol (classes/functions/enums.. etc) and talks to other agents (code-symbols) via natural language. How does the agent even know which code-symbol to talk to? Well since we have access to the editor, the agent can literally perform clicks